Study: best-of-100 voice-acting selection across speaker identities.
Model: an 8B autoregressive expressive TTS model, generated WITH reference-voice conditioning
(audio temperature 1.0, repetition penalty 1.1, text temperature 0.7, top-p 0.95, top-k 25, bf16).
Reference source: a large open crowd-sourced corpus of expressive human speech (one distinct
speaker clip per group, diverse emotions). Reward: combined = invWER x (norm_genu + norm_blend),
where genu is a commercial-voice genuineness score (0-6) and blend an expressive-quality score (0-10),
each min-max normalized across all generated clips; invWER = max(0, 1-WER) from multilingual ASR.
Achieved throughput: 2.87 clips/s on a single GPU.
English script: You actually think your little apology means anything to me now? I have met more honest people crawling in the gutter than I ever found in your company. Do not ever mistake my silence for respect, because you have earned none of it.
German script (translation): Du glaubst tatsaechlich, deine kleine Entschuldigung wuerde mir jetzt noch irgendetwas bedeuten? Ich habe ehrlichere Menschen in der Gosse getroffen, als ich je in deiner Gesellschaft gefunden habe. Verwechsle mein Schweigen niemals mit Respekt, denn den hast du dir in keiner Weise verdient.
Delivery instruction:Cold contempt and disdain, the voice low and sneering, dripping with icy loathing. High quality recording.